_testcapi.sequence_fast_get_item() returns a borrowed reference

[BHUVANSH855]

Crash report

What happened?

_testcapi.sequence_fast_get_item() returns the value from
PySequence_Fast_GET_ITEM() directly.

PySequence_Fast_GET_ITEM() returns a borrowed reference, but the value
returned by the _testcapi function is treated as an owned reference by
the Python caller. As a result, the reference count is not incremented.

I reproduced this on a debug build of CPython 3.15.0rc2.

import sys
import _testcapi

obj = object()
lst = [obj]

before = sys.getrefcount(obj)
result = _testcapi.sequence_fast_get_item(lst, 0)
after = sys.getrefcount(obj)

print("before:", before)
print("after :", after)
print("delta :", after - before)
print("same  :", result is obj)

Affected behaviour

before: 3
after : 3
delta : 0
same  : True

The returned object does not acquire a new reference.

When the returned object is subsequently released, the debug build can abort with:

_Py_NegativeRefcount: Assertion failed: object has negative ref count
Fatal Python error: _PyObject_AssertFailed
Aborted (core dumped)

The expected behaviour is for the returned object to hold a new reference.

Changing the return statement to:

return Py_NewRef(PySequence_Fast_GET_ITEM(obj, index));

makes the reference count increase by one:

before: 3
after : 4
delta : 1
same  : True

I also added a regression test in Lib/test/test_capi/test_abstract.py and verified:

./python -m test test_capi.test_abstract -v

Ran 53 tests
OK

CPython versions tested on:

3.15

Operating systems tested on:

Linux

Output from running 'python -VV' on the command line:

3.15.0rc2 (tags/v3.15.0rc2:435c9e5a798, Sep 2 2026, 13:09:27) [GCC 13.3.0]

Linked PRs

• gh-156831: Fix reference ownership in _testcapi.sequence_fast_get_item #156832

[picnixz]

I do not understand why we should change this? where is it used incorrectly? testcapi is an internal module not for the outside world.

[ZeroIntensity]

For the third time now, I ask that you please search for existing issues on our issue tracker before filing more of these artificial cases. It's unfair to our users to spend our time on things that don't affect them, especially when our backlog is so large.